Starknet integrates Chance for AI agent transaction verification
Here’s a problem that’s only going to get worse: AI agents are increasingly managing money on behalf of humans, and nobody has a great answer for what happens when an agent goes rogue. Chance, a verification harness built for exactly this scenario, is now live on Starknet, where it checks agent-proposed transactions against user intent before execution and manages funds through escrow wallets.
How Chance actually works
The verification process follows a three-step pipeline: PROPOSE, VERIFY, SETTLE. An AI agent proposes a transaction, Chance runs the proposal through a separate model that performs intent matching and price checks, and only verified transactions proceed to settlement.
The system offers three integration tiers, each with a different trust assumption. Light mode keeps everything self-custodial, meaning you hold your own keys and Chance simply validates. Full mode adds escrow wallets with an on-chain judge component, creating a 2-of-2 multi-party computation (MPC) setup where neither the agent nor the user can unilaterally move funds. RPC mode operates as a proxy simulation layer, useful for developers who want to test agent behavior before deploying real capital.
The architecture is designed to work across multiple DeFi venues, including perpetuals and token swaps.
Why Starknet specifically
As a Layer 2 network built on STARK proofs, Starknet enables low-cost verifiable computation, which is the core technical requirement for running a verification layer that doesn’t eat into trading profits with gas fees.
STARK proofs allow Chance to anchor verification results on-chain as attestations. Every time the system confirms that a proposed trade matches user intent, that confirmation gets recorded, creating an audit trail that’s both transparent and tamper-resistant.
Chance was part of Starknet’s PROOF accelerator program, specifically Cohort 01, which ran from July through September 2026. The escrow mechanism in Full mode uses a 2-of-2 MPC setup, meaning both parties must sign off before funds move. Neither the AI agent nor the platform can unilaterally access the escrowed assets.
The bigger picture for AI and DeFi
With a smart contract, the code does exactly what it’s written to do, for better or worse. With an AI agent, the behavior is probabilistic. It interprets your instructions and makes judgment calls. Chance sits directly in that gap, inserting a verification step between intent and execution. The on-chain attestation component means that if something does go wrong, there’s a verifiable record of what was proposed, what was approved, and what was executed.
No token has been announced for Chance, and the project appears focused on building infrastructure rather than generating speculative interest.
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
How much CPO demand will Nvidia Rubin Ultra generate? Bernstein: Optical value per Pod will increase fivefold
Bernstein believes that the co-packaged optics (CPO) and fiber demand driven by Nvidia's Rubin Ultra architecture will surge significantly, with the optical content per Pod rising from $1 million to $5.4 million. This will propel the size of Nvidia's optical ecosystem market to $46.8 billion by 2028 (a 91% annual increase). Although the industry faces supply bottlenecks with InP substrates, the outlook remains positive. The institution reiterates its “outperform” rating for Lumentum, Coherent, and Ciena.

UBS Q3 implements a two-way strategy with gold options, heavily invests in technology stocks, and builds positions in SK Hynix (SKHY.US)
UBS Group (UBS.US) has submitted its Q3 holdings report (13F) as of September 30, 2026.

SpaceX (SPCX.US) reportedly attempts “de-Nvidiaization” at the rack level: developing Kessel cabinets to bypass L6 compute boards, purchasing only chips
SemiAnalysis: SpaceX has developed its own Kessel cabinet, bypassing Nvidia's L6 cards to provide more options for manufacturing its own hardware.

$3.7 billion “data center power equipment prepayments” abandoned! What did "the most active buyer of gas turbines", SoftBank, see?
In the second quarter of fiscal year 2026, SoftBank reported a net loss of $7.1 billion, primarily due to a $3.7 billion impairment charge from a $9.5 billion advance payment for power generation equipment for US data centers—of which about $2.9 billion was non-refundable and effectively lost. The CFO admitted that SoftBank placed orders before the project plan was finalized, and is now conducting a comprehensive review of its procurement strategy. "The most aggressive purchaser of gas turbines has now written off the deposit in this way."
